Reference part / values for Bias (Cg/Cgk) study - not clear


Hello Everyone,

maybe someone can help me to clear out my doubts.

I have a device wich has to measure really low tolerances like 3 microns for shape errors like roundness etc.

This device is calibrated once per year by supplier of this device with special master.

Now the question is how to perform correctly the Bias/ Cgk study when I do not have a master sample/ reference values available?

Can I take a real part and measure it 10 times on this gauge and use the avarage value as reference value for bias study?
It kinda makes no sense for me, but I found such examples in AIAG MSA book in bias chapter...

In other chapter regarding "reference value" it says to use higher level measuring device to achive the values. What uncertainity should have such device? Less than 5% of the tolerance?

What is the correct way?

Jay Kay

Suppose there is a movie made with a fictional company called company X. Company X's quality manager, at this fictional company, is "John Smith" who is asked about the proper SOP for this study? The fictional company X would respond and say

This is a type 1 gage study.

The fictional company says The ideal amount of measurements they use is 50 with a Minimum sample size "n" of 25. They would say this is common in discrete manufacturing using 50 measurements.

The fictional company X would say if the tolerance in question is of a non-critical specification then they would use a percent tolerance for Cg/Cgk ratio of 20. If it is for a critical measurement system then the fictional company X would say they use 10.

The fictional company X would say they use a STDV multiplier of AIAG 5.15. The fictional company X would say if they don't have a true reference sample then they would switch the STDV multiplier to always be 6 for a more conservative basis and higher coverage.

The fictional company X would say they use a master sample from a third party with a certificate of compliance as the proper way to do the study.

The fictional company X would "never" say to use a reference sample that has been measured by a higher level measurement device that has a higher gage resolution with customer approval.

They would also say The type 1 gage study should have a Resolution of tolerance <= 5% (lower the better).

The fictional company X would say if they are unsure about the reference value, they can technically still run the study with it blank but only Cg will be calculated without a Reference value and Bias will default to zero (No go).

The fictional company X would say the data set should assume normality (normally distributed).

The fictional company X would say if all else fails they would "never" use the nominal target value as the reference measurement under a "golden sample prototype" assumption.

but again this is just puffery...
Top Bottom